1、1/25Contents1.Preface.22.Significance.23.Performance Indicators.34.Standard Progress.45.Key Technologies for Network Energy Saving.55.1 Network Architecture.55.1.1 SAGIN Architecture.55.1.2 New Distributed RAN Architecture.65.1.3 Wireless Intelligent Cloud Network.95.2 Air Interface Energy Saving Te

2、chnologies.105.2.1 Energy Saving Technologies in Spatial Domain.105.2.2 Energy Saving Technologies in Time Domain.125.2.3 Energy Saving Technologies in Frequency Domain.135.2.4 Energy Saving Technologies in Power Domain.145.2.5 New Air Interface Hardware.145.3 Integration with New Technologies.165.3

3、.1 Integration with AI Technologies.165.3.2 Integration with 6G Air Interface Technologies.195.4 Other Technologies.206.Summary and Outlook.227.References.238.Main Contributors.249.Abbreviations.242/251.PrefaceThis White Paper describes the significance and serious challenges of 6Gnetwork energy sav

4、ing,briefly introduces the energy consumption performanceindicators and the standard progress of 3GPP on network energy saving,andhighlights the key technical solutions for network energy saving in aspects,suchas the network architecture,energy saving technologies for air interfaces,integration with

5、 new technologies,and other technologies.Finally,this WhitePaper summarizes the main content and related conclusions,and looks forward tothe future development trends.2.SignificanceWith the rapid development of global economy and science technology,energyissues are becoming more and more prominent.G

6、lobal carbon dioxide emissionhas increased significantly since 2000.With the rapid increase of carbon dioxidein the air,global temperature rises rapidly,and extreme weather such as stormsand heat waves caused by global warming seriously endangers human life andproperty.In China,the dual carbon goals

7、,that is,peak carbon dioxide emission by 2030and carbon neutralization by 2060 are included in the 14th Five-Year Plan.Thedual carbon goals are major responsibilities of the global to cope with climatechanges and important cornerstones for sustainable development of industries andenterprises.In the

8、telecommunications industry,carbon dioxide emissions aremainly from consumed electricity.The energy consumption of base stations,communication equipment rooms,and data centers accounts for the majorproportion of the total energy consumption.Therefore,it is critical to save energyfor these items.The

9、energy consumption of a 5G base station at full load is about 3to 4 times that of a 4G base station.Especially with the formal commercial use of5G networks,energy consumption increases significantly.The six typical scenarios and fifteen capability indicators1proposed for 6Gpose higher requirements o

10、n speed,capacity,latency,positioning,and userexperience from multiple dimensions such as intelligence,sensing,and ubiquity.This drives 6G to higher frequency,larger bandwidth,and more computing power,which brings severe challenges to 6G network energy saving.I.Higher frequency:The coverage radius of

11、 6G millimeter wave base stationsis only 30%that of 5G 3.5 GHz base stations,and the power amplificationefficiency of 6G millimeter wave base stations is about 7%to 15%.The specificvalue varies depending on the process.For example,it is 7%+for the silicon3/25germanium(SiGe)process and 15%+for the ga

12、llium nitride(GaN)process,whichis only 1/7 to 1/3 that of traditional 5G base stations.Therefore,more energy isrequired to support normal operation of the power amplifier(PA)of 6G millimeterwave base stations.II.Larger bandwidth:Large bandwidth and multiple antennas are the mainfactors for increased

13、 power consumption of 5G base stations.The powerconsumption of a 5G base station is 3 to 4 times that of a 4G base station.According to the intergenerational growth pattern of bandwidth,it is expected thatthe 6G bandwidth can reach 500 MHz to 1 GHz.If the transmit power per unit ofbandwidth remains

14、unchanged,it is estimated that the transmit power of 6G basestations is more than five times that of 5G base stations,and the overall powerconsumption of a 6G base station is more than four times that of a 5G base station.III.More computing power:Endogenous intelligence is an important featureof 6G.

15、Commonly used artificial intelligence(AI)models include a dozen ofmegabytes to hundreds of gigabytes of model parameters.For example,ChatGPTcontains 175 billion model parameters,and uses 10,000 V100 GPUs for modeltraining.According to a rough calculation by the Global Zero Emission ResearchCenter(GZ

16、R),its power consumption exceeds 1.68 million kilowatt-hours.If thenumber of visitors per day is 1 million,about 12,000 kilowatt-hours of electricityare consumed every day.Green and energy saving should be the basic principles for developing newinnovative 6G technologies,so as to improve system ener

17、gy efficiency andimplement a green and ecological operating model.In addition,6G technologiesshould empower thousands of industries to help all walks of life perform digitaltransformation thoroughly,implement green development strategies,and jointlywrite a new chapter of a shared future for the mank

18、ind.3.Performance IndicatorsEnergy efficiency is an important performance indicator for evaluatingnetwork energy consumption.We can find effective methods for network energysaving from the definition of energy efficiency.Energy efficiency is defined in academic research as the amount of data thatcan

19、 be transmitted under unit of energy consumed and is measured in bit/joule(bit/J).To improve energy efficiency,we can consider the amount of datatransmitted and energy consumption.First,we can effectively improve thetransmission rate.Second,we can reduce the energy consumed for transmitting theunit

20、amount of data.4/25In addition,the international telecommunication union(ITU)defines theenergy efficiency of traditional 5G networks as the air interface capabilities2thatare related to provided services and used to minimize energy consumption of theradio access network(RAN).The energy efficiency in

21、cludes two factors:(1)Network side:the number of information bits transmitted or received by the userequipment(UE)on the RAN per unit of energy consumed;(2)UE side:unit ofenergy consumed by communication modules,measured in bit/J3.Therefore,toimprove network energy efficiency and achieve network ene

22、rgy saving,we cantake measures from the network and UE.As 6G applies to diversified application scenarios with different transmissionperformance requirements on air interfaces,the network energy efficiency can bedefined as the ratio of the performance indicator to the power consumption in aspecific

23、scenario.This helps reflect the actual performance requirements andenergy consumption in the scenario more comprehensively and objectively.Specifically,energy efficiency can be defined as the data rate per unit ofenergy consumed in high-speed scenarios,measured in bps/J,the transmissionlatency per u

24、nit of energy consumed in low-latency scenarios,measured in s/J,andthe coverage distance per unit of energy consumed in wide coverage scenarios,measured in m/J.Similarly,in integrated 6G scenarios,the energy efficiency canbe considered in multiple dimensions.The energy efficiency corresponds toperfo

25、rmance indicators in 6G scenarios,which helps evaluate the air interfaceperformance per unit of energy consumed more comprehensively and meetrequirements in diversified 6G application scenarios.4.Standard Progress3GPP,an international mobile communication standard organization,carriedout discussions

26、 on network energy saving technologies in Release 18,whichincludes the study item(SI)and work item(WI).(1)SI StageAt the RAN meeting#94e held in December 2021,the research content ofnetwork energy saving was formally determined,which mainly includes thefollowing three aspects:Establish an energy con

27、sumption simulation model for base stations toevaluate the performance of network energy saving solutions.Provide evaluation methods and KPIs for network energy saving.5/25Research and identify energy saving technologies on the gNB and UEsides.The SI stage lasts for nearly one year,focuses on the si

28、mulation models andevaluation methods for network energy saving,and the network energy savingtechnologies in time,frequency,spatial,and power domains,UE auxiliaryinformation,and other aspects,and provides the simulation results.Relatedresearch content forms TR 38.8644.(2)WI StageBased on the researc

29、h progress in the SI stage,the WI stage focuses on sometechnologies with high energy saving gains.At the RAN meeting#98 held inDecember 2022,the work content of network energy saving was determined asfollows:Network energy saving technologies in spatial and power domains basedon channel state inform

30、ation(CSI)enhancementCell discontinuous transmission/reception(DTX/DRX)SSB-less SCell in inter-band multi-carrier(CA)scenarios(applicable onlyto FR1 and co-located cells)Solution for preventing legacy UEs from camping in cells implementingRel-18 network energy saving technologiesInter-node beam acti

31、vation and enhanced paging within limited areasEnhanced cell handover processSome network energy saving technologies studied in the SI stage are notstandardized.To furtherreducenetworkenergyconsumptionandlaythefoundation for 6G,3GPP Release 19 further standardizes network energy saving.5.Key Technol

32、ogies for Network Energy Saving5.1Network Architecture5.1.1SAGIN ArchitectureFuture networks will realize the Internet of Everything(IoE),and more naturalspaces,such as space,air,ground,and ocean will be covered,ensuring ubiquitousconnectivity in all domains.As digitalization is accelerated in all w

33、alks of life,digital infrastructure in all domains will be expanded rapidly,and the contradictionbetween the development ofdigital economyandthe increase of energyconsumption and carbon dioxide emission will become increasingly prominent.6/25Green and energy saving will become endogenous demands for

34、 future 6G networkarchitecture.To achieve the development goals of IoE,green,and low carbon,the6G network architecture will undergo a subversive reconstruction.The legacyborderedandchimney-likeRANarchitecturewillbeconvertedtothespace-air-ground integrated network(SAGIN)architecture featuring ubiquit

35、ous,green,and energy saving.The SAGIN architecture consists of three layers:satellite network,airbornenetwork,and low-altitude and ground network,forming a three-dimensionalall-domain coverage network based on terrestrial networks and expanded bynon-terrestrialnetworks.Theterrestrialandnon-terrestri

36、alnetworksareinterconnected and deeply integrated,use a unified protocol stack,and supportperception-free,simplified,and ubiquitous access of massive volume of users.Theterrestrial and non-terrestrial networks can use new networking modes such assuper cellular and non-cellular that are in line with

37、the development trend of greencommunication.In the super cellular architecture,the control plane and user planeof a base station are decoupled,and control and service base stations can bedeployed independently as required.The control base station connects to UEs,transmits control signals,and can ado

38、pt the large area coverage mode.The servicebase station provides users with high-speed data transmission and can be flexiblydeployed as required.Multiple service base stations can be deployed within thecoverage of a control base station,and the service base stations can dynamicallysleep based on cha

39、nges in service load.In this architecture,network coverage canbe dynamically adjusted based on service requirements.When the coverageperformance is not affected,service base stations can sleep at appropriate time,ensuringmoreflexiblesleepandimprovingnetworkenergysaving.Thenon-cellular architecture i

40、s UE-centered,with multiple distributed access points(APs)and a centralized unit(CU)connected to all APs deployed.Throughcentralized signal processing of the CU,widely distributed APs can achievehigh-level collaboration and form a super base station that covers the entire area.Each UE accesses a spe

41、cific group of APs.Spatial macrodiversity and low pathloss can be used to improve the spectral efficiency and energy efficiency of thenetwork.When there are few users in an area,some APs can be shut down tofurther reduce system energy consumption56.5.1.2New Distributed RAN ArchitectureTo better supp

42、ort vertical industries such as autonomous driving,intelligentmanufacturing,andtelemedicine,therearehigherlatencyandreliabilityrequirements.Especially for ubiquitous connections of future 6G networks,the7/25traditionalcentralizedintelligentnetworkarchitecturecannotmeettherequirements.Therefore,the i

43、ndustry has proposed a series of new distributedRAN architectures,which introduce a distributed intelligent computing frameworkto fully utilize the multi-dimensional data and computing resources held by UEsand nodes.However,the new distributed RAN architecture faces many challenges,such as continuou

44、s expansion of distributed nodes,transmission of massivevolume of high-dimensional model parameters,and super computing power.As aresult,6G network energy consumption has become one of the main bottlenecksfor its large-scale deployment and widespread application.The distributed,hierarchical,and inte

45、lligent RAN architecture design is adopted to effectivelyreduce the energy consumption of the 6G distributed RAN7.As the RAN adopts a multi-layer network topology,intelligent functionalcomponents can be deployed at different layers,such as macro/micro base stations,CU/DU,cloud/edge to carry out wire

46、less distributed learning.From bottom to top,the architecture consists of the intelligent UE layer,the first intelligence layerdeployed on the DU,the second intelligence layer deployed on the CU,the thirdintelligence layer deployed on the edge node,and the fourth intelligence layerdeployed on the cl

47、oud,as shown in the following figure.Different intelligentlayers generate different functional configurations for different goals,building adistributed,hierarchical,and intelligent RAN architecture for 6G networks.On thenetwork,intelligentfunctionalcomponentscanbeflexiblyandquicklyorchestrated and u

48、sed,multi-level data analysis network elements can be deployed,and a distributed collaborative control system can be formed at all network layersto achieve distributed intelligent interaction and collaboration between wirelessnodes horizontally8.8/25Figure 1 Schematic Diagram of Hierarchical Deploym

49、ent of IntelligentFunctional Components of a 6G NetworkIn the new distributed RAN architecture,federated learning(FL),as the mostpromising distributed intelligent computing framework for 6G infrastructure,canconduct broader machine learning(ML)while protecting user data privacy.It isexpected to play

50、 an important role in 6G intelligent services and applications.Through integration between FL and multi-layer network topology,multi-levelfederated aggregation can be performed.As shown in the following figure,athree-layernetwork(macrobasestation-microbasestation-UE)formsanFL-based,distributed,hiera

51、rchical,and intelligent RAN architecture in the verticaldirection.Federated aggregation can be split into low-level federated aggregationat micro base stations and high-level federated aggregation at macro base stations.This ensures low communication costs and wider data sharing.Figure 2 Schematic D

52、iagram of FL Node Deployment at Multiple Layers of a6G NetworkSpecifically,ontheedgenetwork,earlymodelaggregationhaslowcommunication costs and can effectively alleviate uncertain model updates due torandom local data.Later model aggregation performs model aggregation andupdate through high-level FL

53、servers,achieving more and wider data sharing andaccelerating convergence of global aggregation.Therefore,federated aggregationcan be flexibly deployed at different network layers based on the actual needs ofthe network for learning performance,latency,capacity,energy consumption,andother indicators

54、 to achieve dynamic multi-level FL.To better save network energy,the communication frequency of high-levelglobal aggregation with high communication costs can be reduced.This reducesthe communication overhead,so as to reduce the total energy consumption.Compared with the traditional FL solution,the

55、distributed,hierarchical,and9/25intelligent RAN architecture that introduces multi-level federated aggregation caneffectively reduce network energy consumption under the same learning accuracy.5.1.3Wireless Intelligent Cloud NetworkThe IMT-2030 Framework9highlights the importance of environmentalada

56、ptability and energy saving and emission reduction of networks and UEs.Network energy efficiency is a quantitative indicator that attracts the mostattention,which is measured in bit/Joule.Around 2010,major operators in China shared their observations on energyconsumption of mobile networks.It was fo

57、und that half of the energy consumptionis from air conditioners and other facilities,and centralized RAN devices wereproposed to reduce energy consumption,which is called cloud RAN(C-RAN).Thefocus can be on the radio or baseband processing,as shown in the followingfigure.Figure 3 C-RAN Schematic Dia

58、gramAccording to analysis1011,C-RAN is eco-friendly infrastructure.First,centralized processing of the C-RAN architecture can exponentially reduce thenumber of base stations and significantly reduce power consumption of on-sitesupportdevices,suchasairconditioners.Second,thecooperativeradiotechnology

59、 can reduce interference between remote radio heads(RRHs)and allowdenser RRHs.Therefore,the distance from RRHs to UEs can be shortened,andsmaller cells with lower transmit power can be deployed without affecting networkcoverage quality.The energy required for signal transmission will be reduced.This

60、helps reduce power consumption in the RAN and extend the battery standby timeof UEs.Finally,the baseband unit(BBU)pool is a shared resource among a largenumber of virtual base stations.Therefore,higher resource utilization and lowerpower consumption can be achieved.When a virtual base station is idl

61、e at nightand does not require most of its processing capabilities,it can be shut down orenter the lower-power state,which does not affect the 24/7 service commitment.Open RAN(O-RAN)inherits the advantages of C-RAN and defines open10/25interfaces of O-CU/O-DU/O-RU.This ensures flexible analysis of t

62、he powerconsumption of different network entities.The report12points out that for mostmobile networks,more than 80%of energy consumption is from the RAN,withthe rest from the core network,support systems,and related cloud infrastructure.Itis estimated that of the 80%energy consumed on the RAN,approx

63、imately 80%ofit is used to power radios,and the remaining 20%of it is used for distributed units(DUs).Radioenergyconsumptioncanbegreatlyreducedbyusingnewtechnologies,such as Micro Sleep Tx and multi-band radio device design andintegration.In addition,with a high-performance general-purpose processor

64、,thecloud-based DU consumes less energy and is compatible with the softwaredevelopment ecosystem.To realize its full potential,the SMO/RIC software iscarefully designed,and the rAPP meets the automated and non-real-time networkmanagement requirements.This helps reduce operating costs,improve network

65、performance,and reduce energy consumption.3GPP carried out discussions and standard work on network energy savingtechnologies in Release 18 and Release 19.For details,see Chapter 4.3GPP pointsout the potential directions for reducing RAN energy consumption.The O-RANarchitecture separates network ent

66、ities and programmable rAPPs,maximizingflexibility.Integrating new energy saving features,the O-RAN architecture isexpected to bring bright prospects in the 6G era.5.2Air Interface Energy Saving Technologies5.2.1Energy Saving Technologies in Spatial DomainThe power consumption of the Active Antenna

67、Unit(AAU)accounts for about80%that of an NR base station and is the main component of network energyconsumption.With tolerable system performance loss,spatial domain energysaving technologies can adapt to spatial elements on the network to significantlyreduce network energy consumption.Depending on

68、the granularity,spatialelements may include antenna elements,TxRUs,antenna ports,antenna panels,and transmission and reception points(TRPs).Compared with semi-static spatialelement adaptation,dynamic spatial element adaptation ensures finer adaptationgranularity,better matches the service load and a

69、ctual transmission environment,and better serves UEs,therefore reducing energy consumption of networks.Massive multiple-input multiple-output(MIMO)is widely used on existing 5Gnetworks because it supports spatial domain multiplexing or multipath diversity.Even though massive MIMO brings large capaci

70、ty,it also increases powerconsumption of base stations due to its large number of TxRUs and relatedhardware processing units(including PAs).An effective network energy saving11/25solution turns on/off TxRUs of base stations based on the traffic load or thenumber of UEs served.As shown in the followi

71、ng figure,when the number of UEsin a cell becomes small,some TxRUs of base stations can be turned off so that thepower consumption of base stations can be reduced without affecting the capacity.Legacy spatial domain shutdown solutions cannot shut down TxRUs dynamicallyand quickly.For example,there i

72、s obvious capacity loss when power saving gainsare obtained,or the solutions cannot be used around the clock because theswitching time for static shutdown is too long.As a result,the antenna status of abase station cannot be quickly adjusted based on the channel status,that is,theantenna status does

73、 not match the channel status,resulting in a great performanceloss.Figure 4 Adaptive and Dynamic TxRU ShutdownBased on the dynamic load and multiple sets of CSI corresponding to differentTxRU shutdown modes,the entire time is divided into many millisecond-levelscheduling windows.Advanced dynamic shu

74、tdown uses a dynamic scheduler tomaximize energy saving gains with limited capacity loss based on energyefficiency.In each scheduling window,all shutdown modes(including noshutdown)are simulated and traversed,so as to quickly select the optimalshutdown mode.In the dynamic shutdown solution,the hardw

75、are response time inenergy-saving state is a key factor affecting network indicators and userexperience.The hardware response time needs to be shortened from minute-levelto millisecond-level,so that energy is saved all the time instead of idle time only.In addition,the number of high-level configura

76、tion parameters for CSI reports islimited.If too many CSI reports are configured for adaptive channel shutdown,theconfiguration of CSI reports for other uses may be affected.Therefore,animportant research direction in the dynamic TxRU shutdown solution is CSI reportenhancement that is,reporting mult

77、iple sets of CSI in one CSI report and reducingCSI overhead.Channel shutdown can reduce the power consumption of the PA,and staticpower consumption of the radio frequency(RF)channel.Spatial domain energysaving technologies based on channel shutdown have obvious advantages inensuring service continui

78、ty,are not limited to base stations with low service load,and are regarded as prevailing solutions for spatial domain energy saving.12/25Large-scale distributed antennas support multi-TRP.That is,multiple antennapanelscanberegardedasapartoftheAAUinspatialdomain.Semi-static/dynamic adaptive multi-TRP

79、 adjustment is a special case of adaptivetransmission antenna adjustment.In actual transmission,a better transmission link(for example,closer physical distance)may exist between a UE and a TRP.Asusing multiple TRPs to transmit data for the UE is not always necessary,dynamically shutting down some TR

80、Ps can significantly reduce the networkenergy consumption.Especially in future transmission systems that deploydistributed massive MIMO arrays to support short-range transmission,more TRPsare used for transmission.Dynamic multi-TRP shutdown can better balance UEperformance and network energy consump

81、tion.5.2.2Energy Saving Technologies in Time DomainIn medium and low system load scenarios,methods such as dynamic andintelligent cell shutdown,intelligentDTX/DRX coordination,andadaptivereduction of broadcast signal transmission time are used to ensure that the networkspends more time in micro,ligh

82、t,and deep sleep states.This effectively reducesthe proportion of signal transmission in time domain,thereby significantlyreducing network power consumption with limited impact on UE performance.When the system load is low,the base station turns off cell DTX/DRX or isshut down for a short period of

83、time,such as shutting down a timeslot,symbol,orsubframe,so that the base station enters the sleep state.Therefore,the energyconsumption of the base station is reduced by prolonging the base station sleeptime and reducing continuous startup or frequent wakeup of the base station.Adaptive adjustment o

84、f the transmission pattern,transmission cycle,and timedomain resource location of common signals or channels,such as SSB,SIB1,paging,and random access channels reduces periodic transmission or reception ofthe base station in always on state and allows the base station to enter a sleep state,thereby

85、reducing energy consumption of the base station.The SSB/SIB-less technology allows a UE to synchronize data and obtainsystemmessagesbasedontheSSB/SIB1transmittedinanothercellinintra-band/inter-band scenarios.In this case,the network serving the UE does notneed to send the SSB/SIB1 on the local carri

86、er to the UE,and therefore the powerconsumption of the network is reduced.In addition,network can send theSSB/SIB1 as required.This reduces transmission of unnecessary SSB/SIB1 andother signals,thereby obtaining network energy saving gains.13/25Cells can be shut down dynamically as required.That is,

87、all or mostcomponents of a cell are shut down as required.Only the sent/received referencesignals,wakeup signals,or discovery signals are retained,minimizing energyconsumed.When a UE within the coverage area of a shutdown cell needs services,the shutdown cell can be activated through cell coordinati

88、on or a wakeup orcommon signal sent by the UE.This balances network energy saving gains andservice performance of the UE.In addition to periodic DTX/DRX on or off,the DTX/DRX mechanism of abase station can use AI to predict the system load or the on/off time of DTX/DRXof the base station.This ensure

89、s intelligent and dynamic DTX/DRX of the basestation and further reduces the energy consumption of the base station.Forcommon signals or channels,AI can be used to predict the UE access status andservices in a cell.With the common signal or channel adaptation mechanism,thebase station intelligently

90、adjusts the transmission of common signals or channelsto reduce the energy consumption.To ensure network energy gains and serviceperformance of UEs during cell shutdown,the key is the time when a cell isturning on or off.AI can predict the system load or the on/off time of a cell,so thatcells can au

91、tomatically shut down at accurate time to further reduce the energyconsumption of the base station.5.2.3Energy Saving Technologies in Frequency DomainFrequency domain energy saving technologies can be classified into two types.In case of a single carrier,the bandwidth can be flexibly adjusted to sav

92、e networkenergy.In case of multiple carriers,some public signals do not need to betransmitted through multi-carrier coordination,thereby reducing network energyconsumption2.In case of a single carrier,6G ultra-large-scale antennas may work in a higherfrequency band and apply to larger bandwidth,whic

93、h inevitably leads to highernetwork energy consumption.On actual networks,the bandwidth part(BWP)allocated to a UE is often higher than the bandwidth actually required by the UE.Network energy consumption can be effectively reduced through efficient BWPadjustment.For example,multiple alternative BWP

94、 resources are allocated to a UEat the same time,and then an appropriate BWP is selected based on actual serviceneeds.In case of multiple carriers,coordination among the carriers plays animportant role in network energy saving.For 5G networks,multi-carrier networkdeployment has become a typical case

95、 for effectively increasing the overall14/25network capacity.In the 6G era,spectrum resources are limited,and frequencyaggregation among and within carriers will become more common.In case ofinter-carrier aggregation,a reference carrier can be used to achieve multi-carriercoordination.The base stati

96、on transmits public signals sent by a UE for randomaccess and related information of the target carrier on the reference carrier.TheUE only needs to detect the data on the reference carrier to flexibly access anoperating carrier.Multiple UEs can share the same reference carrier,and publicsignals for

97、 random access do not need to be sent on their operating carriers,thussaving network energy.5.2.4Energy Saving Technologies in Power DomainFor networks with heavy load,there are usually no idle resources that can beshut down.To reduce network energy consumption,the transmit power needs to bereduced,

98、which can be achieved through dynamic power spectral density(PSD)adjustment.According to some field test results,halving the transmit power in fullload scenarios can reduce the network energy consumption by about 20%,whilehaving little impact on cell throughput.Based on the existing protocol and dev

99、iceimplementations,adjusting the antenna transmit power often needs to reconfigurenetwork parameters,which is inconvenient to operate.The antenna architectureand related power adjustment mechanisms can be optimized so that the networkpower can be adjusted quickly to achieve energy saving.Power domai

100、n energy saving technologies further include PA performanceoptimization and hardware implementation.For details,see Chapter 5.2.5 NewAir Interface Hardware.5.2.5New Air Interface HardwareUltra-Integrated 6G ChipThe 6G base station system needs 1 Tbps data rate for data receiving anddecoding,which is

101、 ten to a hundred times that of 5G base station system.Therefore,we need a ultra-high-speed baseband chip for baseband processing.Channel coding has the highest requirements on baseband chip performance,which determines the power consumption,maximum throughput,complexity,andcosts of the baseband chi

102、p.Therefore,we may need to redesign channel codingalgorithms and standards to improve efficiency or the possibility of parallelprocessing.The 6G base station system needs to handle peak throughput of Tbps15/25level in real time,and therefore has stricter requirements on chip performance.6Gbase stati

103、onswill use dedicated integrated circuit chips toreduce powerconsumption and improve computing efficiency.Fortunately,the chip processevolves continuously,and the integration level is improved continuously.Smallerprocessnodesindicatesmallertransistors,fasterspeeds,andlessenergyconsumption.Ultra-inte

104、grated chips are expected to be commercially used on alarge scale.For example,the baseband chip process can reach 2 nanometers.Figure 5 Technical Roadmap for Process Nodes of a Mainstream FabWith the ultra-integrated baseband and digital intermediate frequency(DIF)chips,as well as the ultra-integrat

105、ed analog RF frontend chips,circuit design andrelated algorithms are optimized continuously to reduce the overall powerconsumption of 6G base stations.This ensures optimal energy consumptionregardless of no load or full load.Efficient 6G PAOnly after being amplified by a PA,RF signals can obtain suf

106、ficient RF powerand be sent out.Statistics show that power consumption of more than half of basestations comes from the PA.The energy efficiency and linearity of the PA arecontradictory to each other.To balance the two,the industry adopts the digitalpredistortion(DPD)algorithm.Digital domain nonline

107、ar filtering opposite to thePA response is performed on the input side of the PA.To continuously optimize algorithms for DPD chips in the 6G base stationsystem,new algorithms applicable to sub-terahertz,terahertz,and gigahertz needto be designed.Since 6G supports extensive frequency bands,multi-band

108、 DPDalgorithms need to be studied.The 6G communication system will use very highcarrier frequencies,so beamforming is used to avoid high attenuation and path loss.Phased array beamformers usually apply multiple PAs to a single digital stream,which brings new DPD issues because multiple parallel PAs

109、must be linearizedusing a single DPD chip.In short,the next-generation DPD algorithm of DIF chipswill directly affect the energy efficiency of 6G base stations.With significant bandwidth increase,the Doherty architecture of the PA isaffected by broadband characteristics and has decreased performance

110、.However,16/25the architecture is simple and has low costs.The outphasing and LMBAarchitectures have better broadband characteristics,but require two RF channels.The architectures are complex and have high costs.In addition,LMBA willintroduce bridge insertion loss,so it has no absolute advantage.As

111、the bandwidthincreases,the base station PA architecture gradually evolves from the singleDohertyarchitecturetocoexistenceofmultiplearchitectures,andthe PAarchitecture is selected based on the scenario.GaN has obvious advantages in power and efficiency for bandwidth lower than100 GHz.Therefore,GaN PA

112、 transistors are widely used.Bandwidth higher than100 GHz is very attractive to 6G,but brings challenges to many major devices,especially the PA and low noise amplifier(LNA).SiGe plays an important role inthe sub-terahertz range of 100 to 300 GHz.For some applications requiring hightransmit power,th

113、e above-mentioned new compound semiconductor materials canbe used at the front end if it is economically feasible.In the range of 300 GHz to 1THz,new compound semiconductor materials such as indium phosphide(InP)feature high speed and can be used for PAs.SiGe and InP ensure lower PAefficiency than G

114、aN.It is necessary to design new waveforms with low peak toaverage power ratio(PAPR)or adopt some technologies to reduce PAPR.Regardless of whether low-order modulation or high-order modulation is used,low PAPR is required.We can reduce power backoff and improve efficiency of thePA to make up for ma

115、terial defects.5.3Integration with New Technologies5.3.1Integration with AI TechnologiesWith the rapid development of AI technologies,we can integrate the AItechnologies with energy saving technologies to intelligently analyze resource andenergy consumption based on actual service prediction,manage

116、network resourcesin a unified manner,and achieve AI-based green and energy-saving digitalization.Based on the 6G network structure,service characteristics,and transmissionenvironment,intelligent and dynamic adaptation of the antennas,panels,TRPs,and RF channels on the base station side can significa

117、ntly reduce the number ofRF chains activated on the network side and network power consumption.Forexample,in an AI-enabled network architecture,AI can be used to dynamicallypredict the shut-down of antennas,e.g.,TRPs,RF channels,antenna ports,andantenna elements.The CSI report enhancement technology

118、 effectively reduces theamount of CSI feedback through MIMO enhancement.Alternatively,throughAI-based channel status learning and training,channel status indication parameters17/25are effectively obtained and compressed,so that more spatial domain adaptationmodes are supported.This promotes network

119、energy saving.(1)AI-basedchannelestimationenhancement:Forhigh-dimensionalchannel matrices on 6G networks,traditional DFT-based type I/II codebooks maynot be able to perform CSI quantification and feedback efficiently underacceptable resource overhead.After channel estimation is integrated with AItec

120、hnologies,powerful learning capabilities of the neural network assist in channelinformation measurement and feedback.This provides a new idea for efficientlyand accurately obtaining channel information and reducing pilot overhead.Inaddition,differentsizesoffeedbackoverheadareusedfortrainingandoptimi

121、zation based on actual needs in different scenarios,ensuring channelfeedback accuracy.(2)Intelligent beam management:For future 6G networks,beams will bemore refined and more beams will be used.To reduce latency and overhead causedby processes such as beam scanning,measurement,and feedback,we can at

122、temptto integrate beam management with AI technologies to achieve intelligent beammanagement and improve beam management efficiency and performance.(3)Adaptive multi-panel communication:In the ultra-large-scale antennasystem of 6G networks,devices are more diversified and both base stations andUEs m

123、ay be equipped with multi-panel.Based on the channel status and systemrequirements,the network side can flexibly schedule different antenna panels forcooperative or non-cooperative transmission,thereby improving system capacityand robustness.To select and schedule panels for transmission,we can use

124、AItechnologies for adaptive multi-panel communication.This further improvessystem performance.In traditional energy saving solutions,a preset threshold is used to triggerenergy saving.After energy saving is triggered,the base station shuts down arelated symbol,channel,cell,or carrier,enters deep sle

125、ep,or performs otheroperations to save network energy.However,the traditional energy savingsolutions have a bottleneck in configuring the energy saving trigger conditionsreasonably and accurately based on the massive volume of data because the datacollection and configuration workload is huge.Theref

126、ore,we need to sense bothenergy efficiency information and service experience to generate more targeteddata and meaningful feedback and reduce mismatches.Thecoreof AItechnologiesliesinfittingthelinearandnonlinearrelationships of data characteristics.ML-based AI technologies mainly rely ondata and mo

127、dels.For data,we need to provide functions to collect,store,access,and share data collected from air interfaces and data related to air interface18/25operations.For models,we need to design architectures and deploy resources tosupport model training,update,access,sharing,and deployment.We can useML-

128、based AI technologies to predict load and service of quality(QoS)of basestations,and associate the ML-based AI technologies with energy saving solutionsto break the above bottlenecks and improve network energy efficiency.Communication services have obvious tidal effects.With historical data andreal-

129、time traffic information,AI technologies can predict traffic accurately andturn on or off base stations dynamically to meet traffic requirements in differenttime periods.Therefore,a new network energy saving strategy13based on AIscene recognition and traffic prediction is proposed,which mainly inclu

130、des threemodules:base station scene recognition based on natural language processing(NLP),AI-based network traffic prediction,and intelligent shutdown based onbase station load and network requirements.First,after NLP of a base station name,the specific scene,such as school,office building,hotel,sho

131、pping mall,stadium,and scenic spot of the base stationcan be accurately identified.A pre-trained general traffic prediction model isprovided for each scene,which provides a basis for subsequent energy savingdecisions.The models include long-and short-term time-series network(LSTNet),neuralbasisexpan

132、sionanalysisforinterpretabletimeseriesforecasting(N-BEATS),and temporal convolutional network(TCN).The traffic predictionmodel predicts network status indicators of a cell in the next period,such as theaverage number of RRC connections,user-plane traffic of uplink and downlinkPDCP layers,and uplink

133、and downlink PRB utilization.Based on weightedcalculation of these indicators,load of the cell in the next period is obtained,which is used to evaluate the working status and load bearing capacity of the basestation.Base stations whose load is lower than a specific threshold are listed ascandidate b

134、ase stations to be shut down.In the area where each candidate basestation is located,coverage of other base stations is calculated to determinewhethershuttingdownthecandidatebasestationhasgreaterimpactoncommunication of UEs within the coverage area.Specifically,the coverage areasof surrounding base

135、stations are calculated.If the overlapping coverage areabetween the candidate base station and other base stations reaches a specificthreshold,and the load of surrounding base stations does not exceed the upperlimit after the candidate base station is shut down,the candidate base station willbe shut

136、 down to save network energy1415.Intelligentnetworkenergysavingsolutionspreventdigitalislandsofinfrastructure,collect precise network energy consumption data in real time,andintelligently manage and control basic resources in equipment rooms such aspower supplies and backup power.In addition,the ene

137、rgy saving strategies for19/25networks,sites,and devices can ensure intelligent collaboration among networksin different modes,and differentiated scheduling of one strategy for one site,achieving user perception-free,high network energy efficiency,and low operationand maintenance(O&M)costs1617.5.3.2

138、Integration with 6G Air Interface TechnologiesWe can integrate network energy saving technologies with 6G new airinterface technologies to reduce network energy.(1)Sensing-based network energy saving:The 6G sensing technology cansense the wireless transmission environment,as well as the UE location

139、anddensity.It helps the base station turn on/off TRPs,antenna ports,antenna elements,and transmission beams.Future integrated sensing and communication systemscan make full use of communication network advantages to provide valuableservices,such as positioning,imaging,and virtual environment reconst

140、ruction.These services can further improve mobile communication performance.Forexample,the sensing technology significantly improves energy efficiency of theentire wireless system18.Depending on different application scenarios and use cases of sensing,informationtobesensedmayincludemotionmonitoring,

141、environmentalmonitoring,and target object detection and tracking.The sensed information canbe used to make energy saving decisions for various communication networkelements(NEs)in the wireless system.Alternatively,the sensed information isused by network management functions to improve energy effici

142、ency of the entirewireless system.Specifically,the sensed motion monitoring information cantrigger UEs or wearable devices to change their activity statuses,for example,entering the power saving mode or long-term monitoring state.On the contrary,itenables the devices to enter the active state.The se

143、nsed environmental monitoringinformation is used to broadcast alarm information in an accurate affected area.This prevents large-scale transmission of redundant signaling.The sensed targetobject detection and tracking information helps allocate or schedule radioresources dynamically in real time bas

144、ed on the information type,so as to improvethe overall transmission efficiency of the wireless system.In addition,theinformation is used to dynamically turn on/off specific resource sets,such asbeams,carrier frequencies,and cells.To achieve energy saving on 6G networks,weshouldestablishinterfacesand

145、protocolstacksbetweentheSensing-as-a-Service(SaaS)and different energy saving NEs(such as basestations,RF units,and UEs),and design energy saving mechanisms or strategies.20/25(2)RIS-based network energy saving:Consisting of low-cost passive ornear-passive components,reconfigurable intelligent surfa

146、ce(RIS)achieves aninnovate antenna structure.It serves as a network node to reflect or transmit beams.This helps significantly improve network coverage with unchanged networkenergy consumption.With 6G RIS,we can use less RF channels and PAs with highconversionefficiencytoreducethenumberofRFchannelsa

147、ndpowerconsumption of base stations.Alternatively,we can use RIS-based relay nodes toreduce network energy consumption.(3)Network energy saving based on enhanced physical layer technologies:Potential new technologies,such as 6G low-density parity check(LDPC),polarcoding,probabilistic shaping modulat

148、ion,scheduling-free non-orthogonal multipleaccess(NOMA),and full duplex can be used at the physical layer to improvesystem spectral efficiency and reduce access overhead.In this case,system energyefficiency is improved without affecting access requirements of all types of UEs inall 6G service scenar

149、ios.5.4Other TechnologiesIntegrated DevicesFirst,we use ultra-wideband RRUs/AAUs,employ advanced algorithms forunified data processing in different frequency bands,integrate multiple modules toat least half the number of RF modules and feeders,and use wideband PAs madeof new materials to achieve ult

150、ra-wide transmission bandwidth on a single channel.In addition,we adopt advanced ultra-wideband DPD architectures and algorithmsto achieve transmission bandwidth of hundreds of megabits per second or evengigabitspersecond,dramaticallysimplifythesitesolution,improvetheintegration level of RF devices,

151、and significantly reduce power consumption.Second,we adopt the super baseband group solution.In this new network model,BBUs are placed in a centralized manner,the RRUs are connected to BBUsremotely through optical fibers,and distributed antennas may be supported.Alarge number of BBUs can be stacked.

152、This ensures dynamic configuration andflexible scheduling of baseband resources and full use of system resources.Multiple BBUs are centrally placed in existing equipment rooms(such as the coreor backbone equipment room).Existing equipment rooms are fully utilized,and nonew equipment room is required

153、.Live network resources,such as varioussupporting facilities and transmission resources in existing equipment rooms areshared to the maximum extent,saving network deployment and maintenance costs.Finally,with the same hardware platform,multiple technical systems can share thesame platform through so

154、ftware configuration.This achieves multi-mode and21/25multi-band integration and smooth network evolution.Green EnergyTo ensure sustainable energy supply,the proportion of green energy in energysupply of 6G networks increases year by year.Green energy includes solar energy,wind power,hydropower,nucl

155、ear power,bioenergy,and geothermal energy.It isexpected that this proportion will exceed 60%by 2035.Green energy usually hasunstable output.For example,wind power is greatly affected by the wind volume,and solar panels cannot generate electricity at night when there is no sunlight.Therefore,an energ

156、y storage system(ESS)is required for peak shaving,valleyfilling,peak regulation,and frequency modulation,and a sensor needs to beembedded in the ESS to collect parameters related to green energy.This helpscomprehensively improve the stability and efficiency of new energy systems,improve clean energy

157、 consumption of 6G devices,reduce the proportion of energywith high carbon dioxide emissions consumed by 6G devices,and directly reducecarbon dioxide emissions of 6G devices at the operation phase.Efficient Data CentersAs the amount of data increases,energy consumption of data centers increasesrapid

158、ly.In traditional centralized computing models,a large amount of data needsto be transmitted from edge devices to remote data centers for processing,and datatransmission usually requires network bandwidth and energy.First,we employdistributed shared computing,dynamically and flexibly deploy required

159、 resources(such as CPU,GPU,NPU,and storage)without being restricted by equipmentrooms,and share computing resources between different servers.Second,we adopta best-effort UE priority mechanism.6G computing tasks,such as XR renderingand AI-based computing are preferentially executed on the UE side.If

160、 the UE sidecannot execute intensive computing tasks,the tasks are pushed to the edge cloudfor execution through the 6G RAN,and then the intensive computing results arereturned to the UE side.If the edge cloud cannot execute the tasks,the tasks arepushed to the central cloud for execution through th

161、e 6G wired network,and theintensive computing results are returned to the UE side.Finally,we take both thecommunication network and computing network into consideration.Based on 6Gwireless channel quality and UE computing capability,computing tasks can bedynamically switched between the edge cloud a

162、nd UE.In summary,the precedingtechnologies help reduce data transmission and processing latency and ensureoptimal energy consumption and computing power allocation in data centers.Network and Power Grid CollaborationThrough coordination between the 6G network and distributed microgrids,we22/25achiev

163、e flexible source-grid-load-storage interaction and coordinated operation ofmassive distributed resources on the edge side,form a flexible sharing mechanismbetween the power and communication networks,realize cross-network resourceoptimization,significantlyreducecarbondioxideemissionsofthe6Gcommunic

164、ation system,and improve the energy and carbon efficiency of the 6Gnetwork and power system.Intelligent Operation ManagementIn termsof refinedoperation,we use intelligentmethods toachieveend-to-end and intelligent control of power consumption,use network energyefficiency management tools to achieve

165、precise power supply and backup,quicklyidentify network energy efficiency bottlenecks,and improve network energyefficiency to significantly reduce operating costs and ensure optimal networkefficiency throughout its lifecycle.Green Supply ChainWe strengthen supply and demand docking,exchange,and coop

166、erationbetween the upstream and downstream of the 6G supply chain,provide moreenergy-saving and low-carbon products,and enhance green capability building andcollaborative cooperation in design,R&D,production,procurement,packaging,logistics,O&M,recycling,and other links,realizing a green industrial c

167、hain.6.Summary and OutlookIn the 5G era,the network energy consumption issue is already prominent.With higher frequency,greater bandwidth,and more computing power,6Gnetworks face severer energy consumption challenges.To achieve the dual carbongoals and ensure sustainable development,we launch the in

168、novation on SAGINarchitecture,new distributed RAN architecture and wireless intelligent cloudnetwork for hybrid networking in multiple scenarios.We implement innovative airinterface energy saving technologies in the spatial,time,frequency,and powerdomains and adopt new air interface hardware to comp

169、rehensively build a digital,networked,and intelligent 6G cloud network energy saving system,and integrateenergy saving technologies with 6G new air interface technologies,such as AI,sensing,and RIS.In addition,we improve device integration,use green energy andefficient data centers,support network a

170、nd grid collaboration,intelligent operationmanagement,and green supply chains,strengthen supply and demand docking,exchange,and cooperation between the upstream and downstream of the 6Gindustrychain,and providemore energy-savingandlow-carbonproducts,realizing a green industrial chain.23/25When study

171、ing energy saving solutions,we also pay attention to the impact ofrelevant solutions on user experience and system performance,and strive to find abalance to meet the 6G vision.We are willing to join hands with colleagues in theindustry to achieve the dual carbon goals and create a green and beautif

172、ul future.We will comprehensively implement the national strategy of dual carbon goals,empower the society with own capabilities,turn green and low-carbon fromchallenges into driving opportunities,and contribute to the dual carbon goals.7.References1DraftNewRecommendationITU-RM.IMT.FRAMEWORKFOR2030

173、ANDBEYOND,Framework and overall objectives of the future development of IMT for2030 and beyond,ITU,2023.2Research Report on Ultra Massive MIMO Technology(Second Edition),IMT-2030(6G)Promotion Group,20223Recommendation ITU-R M.2083-0,IMT-Vision-Framework and overall objectives ofthe future developmen

174、t of IMT for 2020 and beyond.ITU,2015.43GPP TR 38.864 V18.1.0,Study on network energy savings for NR(Release 18),March 2023.5Ni Shunfeng,Luo Jianying,Research on Energy Saving and Emission ReductionTechnologies for Wireless Base Stations J,Communications World,2016(24):61-626China Unicom,6G Green Ne

175、twork White Paper Z,20237Zhao B,Cui Q,Liang S,et al.Green concerns in federated learning over 6GJ.ChinaCommunications,2022,19(3):50-69.8Cuiqimei,Liang Shengyuan,Zhao Borui,et al.,Communication Network ArchitectureGeneration Method,Apparatus,Electronic Device,and Medium P,Chinese patent:CN20211143622

176、7.79ITU-R,Recommendation ITU-R M.2160 Framework and overall objectives of thefuture development of IMT for 2030 and beyond,Nov.2023.10Chih-Lin I,Jinri HUANG,Ran DUAN,Chunfeng CUI,Jesse JIANG,and Lei LI,Recent Progress on C-RAN Centralization and Cloudification,IEEE Access,2014.11CMCC,C-RAN The Road

177、Towards Green RAN.12Ericsson,Open RAN architecture:Embracing energy efficiency.13You Bin,Zhu Wei,Deng Xinhua,et al.,Base Station Traffic Prediction Method,Apparatus,and Electronic Device P,Chinese patent:CN202111647710.X14Lai G,Chang W C,Yang Y,et al.Modeling long-and short-term temporal patterns wi

178、th24/25deep neural networksC/The 41st international ACM SIGIR conference on research&development in information retrieval.2018:95-104.15Oreshkin B N,Carpov D,Chapados N,et al.N-BEATS:Neural basis expansionanalysis for interpretable time series forecastingJ.arxiv preprint arxiv:1905.10437,2019.16Lian

179、g Liwei,Li Rongcong,Research on AI-based Energy Saving Technologies for 5GBase Stations J,Mobile Communications,2019,43(12):32-3617China Unicom,5G Intelligent Energy Saving Technology White Paper Z,201918Research Report on the ISAC Technology,IMT-2030(6G)Promotion Group,20228.Main ContributorsMain c

180、ontributing units and personnel(in random order)S/NContributing UnitContributor1China TelecomJing Guo,Hang Yin2China UnicomLu Li,Jingyan Ma,Wei Zhou3ZTE CorporationJun Xu,Mengzhu Chen4CICT MobileDajun Zhang,Jiaqing Wang5QualcommYiqing Cao,Yan Li6Beijing University of Postsand TelecommunicationsQimei

181、 Cui,Borui Zhao,Yuehan Zhou9.AbbreviationsAAUActive Antenna SystemAIArtificial IntelligenceBWPBandwidth PartCUCentralized UnitCSIChannel State InformationDUDistributed UnitDRXDiscontinuous ReceptionDTXDiscontinuous TransmissionFLFederated Learning25/25LSTNetLong and Short term Time-series NetworkNBE

182、ATSNeural basis expansion analysis for interpretable time series forecastingNLPNatural Language ProcessingPSDpower spectral densityRANRadio Access NetworkRISReconfigurable Intelligent SurfaceRRCRadio Resource ControlSIBSystem Information BlockSSBSS/PBCH BlockTCNTemporal Convolutional NetworkTXRUTransceiver UnitTRPTransmit-Receive PointPDCPPacket Data Convergence ProtocolPRBPhysical Resource BlockUEuser equipment